Beyond the Model: The Edge and Specific Domains Are the Real Battlegrounds

Capabilities that model providers will inevitably absorb should not be treated as durable product moats.

This does not make those features worthless. It means that if a capability naturally becomes part of the foundation model, any product advantage built entirely around it is likely to be temporary.

Prompt completion, intent recognition, and task planning are vertical improvements in model intelligence. Multimodal support across text, images, speech, and video is a horizontal expansion of the model’s boundaries. Both will continue to improve and gradually become general-purpose infrastructure.

The important question is therefore not whether an AI product is ahead today. It is whether the product still makes sense after the next major model upgrade.

The real battleground has two dimensions

Competition beyond the model will not take place only at the edge. It will also take place within specific domains.

The edge determines how AI enters a real-world situation, perceives its environment, and produces action. A specific domain determines whether AI can understand that situation, make reliable judgments, and constrain its behavior.

The former is primarily a tools problem. The latter is primarily a context problem.

Without domain knowledge, an edge system may be able to perceive and act but remain unable to make reliable decisions. Without an appropriate tool system, domain intelligence may remain confined to giving advice in cyberspace.

The edge and specific domains are therefore not two isolated markets. They are two dimensions that an effective AI product must address simultaneously.

The edge provides the tool system

Smartphones will probably remain important interfaces and provide part of the required computing power for a long time. But they do not have to remain the only interface.

Glasses, headphones, vehicles, industrial terminals, and specialized peripherals may all become new entry points.

An interface does more than receive instructions. It determines what an AI system can observe, what it can continuously perceive, and whether it can understand the user’s situation at the right moment.

Beyond interfaces, AI also needs execution tools.

AI should not remain confined to cyberspace. It must eventually affect the physical world: controlling equipment, coordinating resources, performing inspections, assisting operations, or advancing real-world workflows.

This connection should not be reduced to humanoid robots. A robot is a much broader concept than its popular image suggests. Any device, system, or specialized tool capable of receiving an intelligent decision and producing a physical outcome can become part of an AI tool system.

The opportunity at the edge is therefore not merely to create a new device. It is to build a complete tool system embedded in a specific situation, spanning interfaces, continuous perception, and real-world execution.

Specific domains provide high-quality context

An Agent is often described as the combination of:

Model + Tools + Context

The same structure remains valid when expanded to the physical world.

The model provides general intelligence. Edge interfaces, sensing devices, and execution systems provide tools. Knowledge accumulated within specific domains provides high-quality context.

Individual messages, chat histories, and temporary instructions are all forms of context, but they are not equivalent to structured domain knowledge. More context does not necessarily mean better context.

Valuable domain context contains stable concepts, business relationships, decision rules, accumulated experience, and real-world constraints.

It tells an AI system not only what happened, but what matters, why it matters, and how to act under specific conditions.

This is precisely what well-funded model and infrastructure providers cannot exhaustively capture.

They can continue improving general-purpose capabilities, but they cannot automatically acquire the tacit knowledge accumulated inside every domain. This limitation becomes more pronounced in specialized, fragmented, and relatively small niche markets.

The durable moat for AI products

The durable moat for an AI product may not be a stronger model, but two assets that general-purpose models cannot easily absorb:

1. A scenario-specific tool system spanning interfaces, perception, and execution;
2. A continuously accumulated, structured, and difficult-to-copy body of domain context.

Tools determine what an AI system can access, perceive, and change. Domain context determines how it understands the situation, makes judgments, and constrains its actions.

The model sits beneath both as a continuously improving layer of general-purpose infrastructure.

A strong product should not be threatened by model improvements. As models become more capable, the product’s tool system and domain context should become more valuable.

If a model upgrade eliminates a product, that product was probably filling a temporary capability gap. If the upgrade amplifies the product’s advantages in tools and domain context, the product is much more likely to endure.